Online diagnosis of state of health for lithium-ion batteries based on short-term charging profiles

In this study, a machine learning method is proposed for online diagnosis of battery state of health. A prediction model for future voltage profiles is established based on the extreme learning machine algorithm with the short-term charging data. A fixed size least squares-based support vector machine with a mixed kernel function is employed to learn the dependency of state of health on feature variables generated from the charging voltage profile without preprocessing data. The simulated annealing method is employed to search and optimize the key parameters of the fixed size least squares support vector machine and the mixed kernel function. By this manner, the proposed algorithm requires only partial random and discontinuous charging data, enabling practical online diagnosis of state of health. The model training and experimental validation are conducted with different kernel functions, and the influence of voltage range and noise are also investigated. The results indicate that the proposed method can not only maintain the state of health estimation error within 2%, but also improve robustness and reliability. •Online state of health diagnosis is addressed based on short-term charging data.•A mixed kernel function is applied in fixed size least square-support vector machine.•The voltage prediction algorithm is proposed based on extreme learning algorithm.•Experimental results highlight that the proposed method is more accurate and robust.